DESIGNING NOVEL ANTIMICROBIAL PEPTIDE BIOLOGICS: MACHINE LEARNING CLASSIFICATION AND CONSENSUS-SEQUENCE ENGINEERING OF THE BREVININ-2 FAMILY

Colin Milo McDowell

The increasing prevalence and severity of antimicrobial resistant infections are a critical concern for public health systems worldwide. As the race between antibiotic discovery and the emergence of antimicrobial resistance (AMR) continues to escalate, there is an urgent need for novel biologics capable of circumventing classical resistance mechanisms. When taken into consideration alongside the economic and regulatory challenges facing novel antibiotic development, alternative approaches to address bacterial AMR are becoming increasingly popular. Antimicrobial peptides (AMPs), short amphipathic sequences of amino acids found in many innate immune systems, are one potential solution to this issue due to the vast complexity of their sequence space and historical use as broad-spectrum antimicrobial agents. Unfortunately, the extreme size of the sequence space and the logistical challenges surrounding de novo peptide synthesis limit many of the standard approaches to drug development and necessitates an interdisciplinary approach that combines physical chemistry, microbiology, and computational analysis. We investigated the efficacy of a consensus-sequence-driven approach to AMP design and development for the Brevinin-2 AMP family (Chapter 2) by synthesizing four novel peptides (G30, G33, S33, and G37) and one previously characterized peptide (Brevinin-2GUb) and evaluating their activity against a panel of AMR bacteria. We hypothesized that greater conservation of natural residues within subsets of the Brevinin-2 family would result in peptides capable of circumventing classical resistance in both Gram-positive and -negative bacteria. Furthermore, we developed a novel hierarchical ensemble machine learning framework called HierAMP to further explore and strategically refine AMP identification (Chapter 3). The Brevinin-2 family, a large group of peptides secreted from frog skin, exhibits broad-spectrum activity. Herein we employed and validated a consensus-sequence approach to generate a widely applicable chemical peptide synthesis pipeline for the Brevinin-2 family. Peptide G37 exhibited broad-spectrum activity, with the growth of a Class B carbapenemase-resistant Escherichia coli inhibited at 16 µM and a Class A carbapenemase-resistant Klebsiella pneumoniae inhibited at 64 µM (Chapter 2). All peptides displayed limited red blood cell hemolysis and acted in a rapid fashion (Chapter 2). Additionally, when AMP predictions are expanded beyond the family level, using next-generation computational modeling, HierAMP demonstrated extremely high accuracy, precision, sensitivity, and specificity, with an area under the receiver operating characteristic curve value >95% (Chapter 3). Ultimately, the interdisciplinary paradigm established in this work – smooth transition from in-silico design through chemical production toward in vitro evaluation – represents a scalable and systematic approach to AMP development that strengthens the Brevinin-2 family and assists in future drug development.

DESIGNING NOVEL ANTIMICROBIAL PEPTIDE BIOLOGICS: MACHINE LEARNING CLASSIFICATION AND CONSENSUS-SEQUENCE ENGINEERING OF THE BREVININ-2 FAMILY

Files and links (1)

Abstract

Metrics

Details